改进PSO-YOLOv5的金属表面缺陷检测方法

Research on Metal Surface Defect Detection by Improved PSO-YOLOv5

针对已有检测方法在目标尺寸较小时适应度差及特征不清晰导致漏检等问题,提出改进的PSO-YOLOv5金属缺陷检测方法。基于YOLOv5网络结合PSO算法优化网络的颈部部分,同时加入了自适应锚框计算,为边界框提供更准确的移动方向和位置信息。最后使用反向传播神经网络对图像进行训练得到最终的深度卷积神经网络。与传统检测方法相比,改进后的PSO-YOLOv5算法检测速率为31.6frame/s,平均检测精度为78.56%,较YOLOv5提高8.32个百分点,检测精度优于Faster RCNN等算法,表明改进的PSO-YOLOv5算法能够有效提高小目标情况下的缺陷检测精度。

Aiming at the problems that existing detection methods have poor adaptability and unclear features when the target size is small,this paper proposes an improved PSO-YOLOv5 metal defect detection method.This method is based on the YOLOv5 network and combined with the PSO algorithm to optimize the neck part of the network.At the same time,the adaptive anchor box calculation is added to provide more accurate moving direction and more accurate position information for the bounding box.Finally,the back-propagation neural network is used to train the image to obtain the final deep convolutional neural network.Compared with the traditional detection method,the improved PSO-YOLOv5 algorithm has a detection performance of 31.6 frame/s and an average detection accuracy of 78.56%,which is 8.32 percentage points higher than that of YOLOv5.The detection accuracy is better than that of Faster RCNN and other algorithms.The results show that the improved PSO-YOLOv5 can effectively improve the defect detection accuracy in the case of small targets.